Chunk-by-chunk shadowing: learning to speak in breaths, not words
A Chunks tab in the player reveals a sentence one sense-group at a time. You say each chunk out loud before the next appears — training the habit of speaking in chunks instead of translating Korean word by word.
The owner named the pain exactly: "when I speak English I don't speak in chunks, I swap Korean words one at a time." The prepositions work attacked one side of that; this is the other side — output, in chunks.
The clip player already had a blind-shadowing tab (mask the whole transcript to recall it). The new Chunks tab is the complement. It uses the 직독직해 chunks the AI already extracts — sense-groups like so my only goal / with this video / is to have you — and reveals them one at a time. The current chunk is highlighted; the ones ahead are hidden blocks. You say the visible chunk out loud, hit Next, the next chunk appears, and the sentence builds left to right. By the end you've spoken the whole sentence the way a native does — in breaths, not in single-word swaps.
Small choices that matter:
- English first. The Korean gloss is off by default; it's a toggle, not the main event. You're training to produce English, so English leads.
- Self-paced, not timestamped. Chunks have no audio timing, so I didn't fake a synced karaoke. You advance yourself, at speaking speed, with the clip looping in the background for reference. Honest, and actually better for output practice.
- Free to build. It reuses the existing analysis (
chunkedTranslation) — same React Query cache as the analysis panel, no new endpoint, no migration. If a clip hasn't been analyzed yet, it just says so.
With this, both halves of the original frustration have a tool: prepositions (spotlight → study page → drill) and chunked speaking (this). The product is starting to answer the exact problem its owner has.